The present invention relates to the art of diagnostic medical imaging. It finds particular application in conjunction with single photon emission computed tomography (SPECT) and nuclear cameras, and will be described with particular reference thereto. However, it is to be appreciated that the present invention is also amenable to other like applications including, but not limited to, positron emission tomography (PET) and ultrasound tomography.
Attenuation correction is now an integral part of nuclear medicine image reconstruction. Several techniques have been introduced over the years, ranging from simple analytical filters to elaborate transmission devices, to obtain an object--dependent attenuation map. Often, in analytical correction, assumptions are made that are too restrictive to be realistic. Object-dependent attenuation correction based on emission data only is an attractive and theoretically sound alternative, but not ready for large scale utilization. Finally, techniques based on an attenuation map constructed from measured transmission projection data, while a popular approach, suffer from very poor statistical quality. Moreover, current techniques for reconstructing the emission data are sensitive to this noise. Moreover, the collected transmission projection data is often truncated.
The approaches presented herein are a departure from previously available techniques. In order to appreciate the difference, it is helpful to examine previous techniques.
It is usually recognized that iterative methods lend themselves easily to the introduction of additional correction factors such as attenuation. With reference to FIG. 1, the basic concept behind iterative methods, such as maximum likelihood expectation maximization (ML-EM) reconstruction, is to find the object (i.e., image or emission map) for which a mathematical projection produces results comparable to the set of measured emission projections. The ML-EM reconstruction algorithm is one such engine that allows the image to be found in an efficient manner. From an initial guess or assumption, a set of projections is artificially created by a projector employing a projection model. These projections are then compared to the xe2x80x9crealxe2x80x9d or measured set. When certain conditions are met (i.e., when the projections of the emission map are sufficiently close to the measured emission projections), the iterative process stops and the current image is the best possible representation of the object, otherwise, the initial guess is updated and a new set of projections produced. Clearly, the projection model employed is an important part of the projection operation. However, an accurate projection model can be complex and detailed.
With reference to FIG. 2, one way to improve the xe2x80x9crealismxe2x80x9d of the projection model or the update operations is to include a priori information, i.e., what is already known about the object (e.g., its contour, texture, and so forth). In general, this is a powerful tool that offers dramatic improvement of the reconstruction. One caveat is, however, that this information needs to be real. In other words, if a priori information is true or accurate, it helps the reconstruction. On the other hand, inaccuracy tends to inappropriately bias the reconstruction or otherwise introduce unwanted artifacts. Moreover, in practice, it is difficult to find non-trivial characteristics of the object to be imaged that are always accurate.
With reference to FIG. 3, the attenuation map is certainly one element that can help the reconstruction. Information pertaining to attenuation characteristics can be generated artificially for simple situations (e.g, uniform attenuation, symmetrical attenuation, etc.). In general, however, inaccuracies can be introduced by suggesting attenuation features that are not true and/or not object specific.
With reference to FIG. 4, it is significantly more advantageous when the attenuation map is constructed for each object and used directly in the reconstruction. Often, the attenuation map is derived from measured transmission projection data. This is the basis of most of the modern non-uniform attenuation correction devices. However, the problem with this approach is statistics. Typically, in order to define a useful and usable attenuation map, many counts are needed. In practice, however, the definition of the attenuation map is count-limited, and the inherent noise associated with it is transported into the reconstruction of the emission map. Another problem is that often the transmission projection data used to derive the attenuation map is truncated thereby resulting in a degraded attenuation map. In other words, a poor attenuation map can actually degrade the image it aims at improving because of a lack of counts and/or a truncated transmission data set.
These limitations in the acquisition of the measured attenuation maps have provided the motivation for investigation into different techniques. It is generally known that the emission data contains information from which theoretically the attenuation map could be reconstructed. Such reconstruction methods are referred to as sourceless. In fact, each point in the object can be considered as a transmission source, and the observed intensity at any given point on the detector can be compared with the expected intensity without attenuation. Unfortunately, due to limitations, such as the limited number of counts in the emission map, the relationship between the emission and attenuation maps can only partially be established. Moreover, there is an intrinsic inability of the process to differentiate between a low activity, low attenuation condition and a high activity, high attenuation condition.
Yet another drawback of previous techniques is that they tend to employ only a linear projection model. That is to say, they merely account for those events that lie along a singular path or line from the detection bin to the activity. Such projection models are not adapted to account for scatter or collimator resolution which may allow a particular detector bin to detect photons from off-line sources.
The present invention contemplates a new and improved technique for providing attenuation correction in emission computed tomography which overcomes the above-referenced problems and others.
In accordance with one aspect of the present invention, a method of ML-EM image reconstruction is provided for use in connection with a diagnostic imaging apparatus that generates projection data. The method includes collecting projection data including measured emission projection data and measured transmission projection data. Optionally, the measured transmission projection data is truncated. Initial emission and attenuation maps are assumed and iteratively updated. With each iteration, the emission map is recalculated by taking a previous emission map and adjusting it based upon: (i) the measured emission projection data; (ii) a reprojection of the previous emission map which is carried out with a multi-dimensional projection model; and, (iii) a reprojection of the attenuation map. Also, with each iteration, the attenuation map is recalculated by taking a previous attenuation map and adjusting it based upon: (i) the measured emission projection data; (ii) the reprojection of the previous emission map which is carried out with the multi-dimensional projection model; and, (iii) the measured transmission projection data.
In accordance with another aspect of the present invention, a medical diagnostic imaging apparatus for reconstructing image representations of a subject being examined therewith is provided. It includes a emission memory which stores measured emission projection data from the subject. A transmission memory stores measured transmission projection data from the subject. Optionally, the measured transmission projection data is truncated. An image memory stores emission maps, and an attenuation map memory stores attenuation maps. A first projector generates emission map projections via forward projection of the emission maps from the image memory. Likewise, a second projector generates attenuation map projections via forward projection of the attenuation maps from the attenuation map memory. The first and second projectors use first and second multi-dimensional projection models, respectively. A first data processor samples the emission memory, and the first projector, and in accordance therewith updates the attenuation maps stored in the attenuation map memory. A second data processor samples the emission memory, the first projector, and the second projector, and in accordance therewith updates the emission maps stored in the image memory. Finally, a human-viewable display renders reconstructed image representations of the subject from the emission maps.
In accordance with another aspect of the present invention, a method of reconstructing an image is provided for use in connection with a nuclear imaging apparatus. The method includes collecting emission projection data and truncated transmission data. Thereafter, an attenuation map is updated in accordance with the collected transmission projection data, the collected emission projection data, and a reprojection of an emission map. Likewise, the emission map is updated in accordance with the collected emission map, the reprojection of the preceding emission map, and a reprojection of the attenuation map.
One advantage of the present invention is improved attenuation maps for ML-EM reconstruction.
Another advantage of the present invention is multi-dimensional projection modeling which more accurately approximates actual conditions.
Yet another advantage of the present invention is that there is less likelihood of noise being introduced into the final reconstruction via the attenuation map.
Still further advantages and benefits of the present invention will become apparent to those of ordinary skill in the art upon reading and understanding the following detailed description of the preferred embodiments.